The library is now better at handling larger graphs. By increasing the following configuration from their defaults, more shards will be created. This addresses the potential issue that each shard cannot be larger than 1MB (because it is an Entity in NDB).
no_of_subject_shards = 16,
no_of_shards_per_predicate_default = 1,
The first option defines how many shards subjects are divided into (default is 16). A subject is assigned to a shard based on its SHA1 digest. Allowed values for this option is 1, 16, 256, 4096.
The second option defines how many shards each predicate is divided into (default is 1). Triples will be written to a random one of these. Allowed values for this option is 1, 16, 256, 4096.
Predicates will often vary considerable in sizes, so it is also possible to divide a name predicates into more shards than the number set above. This is done by setting the option
no_of_shards_per_predicate_dict = {}
The keys must be URIRefs. Allowed values are 1, 16, 256, 4096. Example configuration:
{'no_of_shards_per_predicate_dict': {URIRef(''http://www.w3.org/2000/01/rdf-schemalabel): 16}